The CredentialProvider API is an SPI framework for plugging in extensible credential providers. Credential providers are used to separate the use of sensitive tokens, secrets and passwords from the details of their storage and management. The ability to choose various storage mechanisms for protecting these credentials allows us to keep such sensitive assets out of clear text, away from prying eyes and potentially to be managed by third party solutions.
This document aims to describe the design of the CredentialProvider API, the out of the box implementations, where they are used and how to adopt their use.
Let’s provide a quick overview of the use of the credential provider framework for protecting passwords or other sensitive tokens in hadoop.
There are certain deployments that are very sensitive to how sensitive tokens like passwords are stored and managed within the cluster. For instance, there may be security best practices and policies in place that require such things to never be stored in clear text, for example. Enterprise deployments may be required to use a preferred solution for managing credentials and we need a way to plug in integrations for them.
There are numerous places within the Hadoop project and ecosystem that can leverage the credential provider API today and the number continues to grow. In general, the usage pattern consists of the same requirements and flow.
Example: ssl.server.keystore.password
hadoop credential create ssl.server.keystore.password -value 123 -provider localjceks://file/home/lmccay/aws.jceks
Note that the alias names are the same as the configuration properties that were used to get the credentials from the Configuration.get method. Reusing these names allows for intuitive migration to the use of credential providers and fall back logic for backward compatibility.
Now, we need to make sure that this provisioned credential store is known at runtime by the Configuration.getPassword method. If there is no credential provider path configuration then getPassword will skip the credential provider API interrogation. So, it is important that the following be configured within core-site.xml or your component’s equivalent.
<property> <name>hadoop.security.credential.provider.path</name> <value>localjceks://file/home/lmccay/aws.jceks</value> <description>Path to interrogate for protected credentials.</description> </property>
A couple additional things to note about the provider path:
In summary, first, provision the credentials into a provider then configure the provider for use by a feature or component and it will often just be picked up through the use of the Configuration.getPassword method.
Feature\Component | Description | Link |
---|---|---|
LDAPGroupsMapping | LDAPGroupsMapping is used to look up the groups for a given user in LDAP. The CredentialProvider API is used to protect the LDAP bind password and those needed for SSL. | LDAP Groups Mapping |
SSL Passwords | FileBasedKeyStoresFactory leverages the credential provider API in order to resolve the SSL related passwords. | TODO |
HDFS | DFSUtil leverages Configuration.getPassword method to use the credential provider API and/or fallback to the clear text value stored in ssl-server.xml. | TODO |
YARN | WebAppUtils uptakes the use of the credential provider API through the new method on Configuration called getPassword. This provides an alternative to storing the passwords in clear text within the ssl-server.xml file while maintaining backward compatibility. | TODO |
AWS S3/S3A |
Uses Configuration.getPassword to get the S3 credentials. They may be resolved through the credential provider API or from the config for backward compatibility. | AWS S3/S3A Usage |
Azure WASB |
Uses Configuration.getPassword to get the WASB credentials. They may be resolved through the credential provider API or from the config for backward compatibility. | Azure WASB Usage |
Azure ADLS |
Uses Configuration.getPassword to get the ADLS credentials. They may be resolved through the credential provider API or from the config for backward compatibility. | Azure ADLS Usage |
Apache Accumulo |
The trace.password property is used by the Tracer to authenticate with Accumulo and persist the traces in the trace table. The credential provider API is used to acquire the trace.password from a provider or from configuration for backward compatibility. | TODO |
Apache Slider |
A capability has been added to Slider to prompt the user for needed passwords and store them using CredentialProvider so they can be retrieved by an app later. | TODO |
Apache Hive |
Protection of the metastore password, SSL related passwords and JDO string password has been added through the use of the Credential Provider API | TODO |
Apache HBase |
The HBase RESTServer is using the new Configuration.getPassword method so that the credential provider API will be checked first then fall back to clear text - when allowed. | TODO |
Apache Oozie |
Protects SSL, email and JDBC passwords using the credential provider API. | TODO |
Apache Ranger |
Protects database, trust and keystore passwords using the credential provider API. | TODO |
Usage: hadoop credential <subcommand> [options]
See the command options detail in the Commands Manual
Utilizing the credential command will often be for provisioning a password or secret to a particular credential store provider. In order to explicitly indicate which provider store to use the -provider option should be used.
Example: hadoop credential create ssl.server.keystore.password -provider jceks://file/tmp/test.jceks
In order to indicate a particular provider type and location, the user must provide the hadoop.security.credential.provider.path configuration element in core-site.xml or use the command line option -provider on each of the credential management commands. This provider path is a comma-separated list of URLs that indicates the type and location of a list of providers that should be consulted. For example, the following path: user:///,jceks://file/tmp/test.jceks,jceks://hdfs@nn1.example.com/my/path/test.jceks indicates that the current user’s credentials file should be consulted through the User Provider, that the local file located at /tmp/test.jceks is a Java Keystore Provider and that the file located within HDFS at nn1.example.com/my/path/test.jceks is also a store for a Java Keystore Provider.
Keystores in Java are generally protected by passwords. The primary method of protection of the keystore-based credential providers are OS level file permissions and any other policy based access protection that may exist for the target filesystem. While the password is not a primary source of protection, it is very important to understand the mechanics required and options available for managing these passwords. It is also very important to understand all the parties that will need access to the password used to protect the keystores in order to consume them at runtime.
Option | Description | Notes |
---|---|---|
Default password | This is a harcoded password of “none”. | This is a hardcoded password in an open source project and as such has obvious disadvantages. However, the mechanics section will show that it is simpler and consequently nearly as secure as the other more complex options. |
Environment variable | HADOOP_CREDSTORE_PASSWORD | This option uses an environment variable to communicate the password that should be used when interrogating all of the keystores that are configured in the hadoop.security.credential.provider.path configuration property. All of the keystore based providers in the path will need to be protected by the same password. |
Password-file | hadoop.security.credstore.java-keystore-provider.password-file | This option uses a “side file” that has its location configured in the hadoop.security.credstore.java-keystore-provider.password-file configuration property to communicate the password that should be used when interrogating all of the keystores that are configured in the hadoop.security.credential.provider.path configuration property. |
Extremely important to consider that all of the runtime consumers of the credential being protected (mapreduce jobs/applications) will need to have access to the password used to protect the keystore providers. Communicating this password can be done a number of ways and they are described in the Options section above.
Keystore Password | Description | Sync Required | Clear Text | File Permissions |
---|---|---|---|---|
Default Password | Hardcoded password is the default. Essentially, when using the default password for all keystore-based credential stores, we are leveraging the file permissions to protect the credential store and the keystore password is just a formality of persisting the keystore. | No | Yes | No (documented) |
Environment Variable | HADOOP_CREDSTORE_PASSWORD Environment variable must be set to the custom password for all keystores that may be configured in the provider path of any process that needs to access credentials from a keystore-based credential provider. There is only one env variable for the entire path of comma separated providers. It is difficult to know the passwords required for each keystore and it is suggested that the same be used for all keystore-based credential providers to avoid this issue. Setting the environment variable will likely require it to be set from a script or some other clear text storage mechanism. Environment variables for running processes are available from various unix commands. | Yes | Yes | No |
Password File | hadoop.security.credstore.java-keystore-provider.password-file configuration property must be set to the location of the “side file” that contains the custom password for all keystores that may be configured in the provider path. Any process that needs to access credentials from a keystore-based credential provider will need to have this configuration property set to the appropriate file location. There is only one password-file for the entire path of comma separated providers. It is difficult to know the passwords required for each keystore and it is therefore suggested that the same be used for all keystore-based credential providers to avoid this issue. Password-files are additional files that need to be managed, store the password in clear text and need file permissions to be set such that only those that need access to them have it. If file permissions are set inappropriately the password to access the keystores is available in clear text. | Yes | Yes | Yes |
The use of the default password means that no additional communication/synchronization to runtime consumers needs to be done. The default password is known but file permissions are the primary protection of the keystore.
When file permissions are thwarted, unlike “side files”, there are no standard tools that can expose the protected credentials - even with the password known. Keytool requires a password that is six characters or more and doesn’t know how to retrieve general secrets from a keystore. It is also limited to PKI keypairs. Editors will not review the secrets stored within the keystore, nor will cat, more or any other standard tools. This is why the keystore providers are better than “side file” storage of credentials.
That said, it is trivial for someone to write code to access the credentials stored within a keystore-based credential provider using the API. Again, when using the default password, the password is merely a formality of persisting the keystore. The only protection is file permissions and OS level access policy.
Users may decide to use a password “side file” to store the password for the keystores themselves and this is supported. It is just really important to be aware of the mechanics required for this level of correctness.